On the Feasibility of Radiomic Analysis for the Detection of Breast Lesions in Speed-of-Sound Images of the Breast

Recent non-linear ultrasound imaging methods estimate acoustic tissue properties, such as speed-of-sound (SOS), density, and compressibility, among others. These methods can be used to generate 2D reconstructions of the properties of inner structures of the breast for further analysis. Due to differences in the acoustic properties between cancerous and normal tissues, these reconstructions are particularly attractive for computerized analysis. In this work, we explored the feasibility of using radiomic analysis on SOS images for breast lesion detection. We performed an in-silico analysis of SOS slices extracted from 120 3D virtual breast phantoms and built a system based on radiomic features extracted from SOS images for the detection of breast masses. We measured the performance of the system in terms of the area under the ROC curve (AUC) with 95% confidence intervals (CI). We also compared the performance of lesion detection from SOS images against a model trained with synthetic mammograms generated from the same breast phantoms. Radiomic analysis on SOS images yielded statistically significant results with AUCs of 0.73 (CI: 0.64-0.82), 0.89 (CI: 0.83-0.95), and 0.94 (CI: 0.89-0.98) at pixel-size of 1.5, 2.0 and 2.5 mm respectively. Radiomic analysis on mammograms showed lower performance with an AUC of 0.62 (CI: 0.52-0.72). Our evidence suggests that the use of SOS images, paired with radiomic analysis, could aid on the detection of breast masses that are hard to recognise using digital mammography. Further investigation on ultrasound-based reconstruction of SOS images of the breast is warranted.